Patients go home. Arogya keeps watching.

The problem

Public hospitals in Thailand and India care for more patients than their staff can follow. The gaps are at the front door, in the consult room, and after the patient goes home.

The solution: one patient, five steps

Arogya follows the same patient from arrival to recovery at home. At each step the work is split: rules and models compute, AI explains and drafts, a person decides.

1. 1

   Arrival: check every patient for red flags

   Nurse · Arrivals

   The nurse types what the patient says and the vitals. Fixed red-flag rules set the level: Emergency, Urgent or Normal. AI writes a short summary and what to check next.

   Human Nurse confirms the level, or changes it with a reason.

   Fixed rule (not AI)AI · LLMHuman
2. 2

   Consult: AI writes, the doctor decides

   Doctor · Consult

   Speech-to-text writes the conversation. AI reads past visits, drafts the note and suggests medicines with dose times, each with a reason. Rx Check rules flag interactions, allergies, duplicates and kidney risk, and block signing until resolved.

   Human Doctor edits, resolves every flag and signs. The risk model then scores the readmission risk.

   AI · VoiceAI · LLMFixed rule (not AI)HumanAI · ML model
3. 3

   At home: the patient asks their own AI

   Patient · Chat

   The patient signs in on their phone and asks the Companion about their medicines, visits, tests and appointments. Answers come only from their own signed record. Emergency words get “call 1669” and alert the ward.

   Human Nurse calls back on anything the record can’t answer.

   AI · LLMFixed rule (not AI)Human
4. 4

   Follow-up: set by the nurse, done by AI

   Nurse · Follow-up

   AI suggests check-in days and writes questions for this patient’s condition. The patient answers in the app; if an answer is unclear, AI asks again with an example.

   Human Nurse edits and approves the plan and picks when the questions go out.

   AI · ML modelAI · LLMHuman
5. 5

   Alert: the right person hears in time

   Nurse + doctor · Alerts

   Fixed protocol rules read the answers (e.g. heart failure: weight up 1.5 kg and swollen ankles) and raise the alert. The risk is updated. AI explains the alert and drafts the next action. Unanswered questions after 24 h become a nurse task.

   Human Nurse or doctor marks it handled, or dismisses it with a reason.

   Fixed rule (not AI)AI · ML modelAI · LLMHuman

Who uses it

Each person has their own account and sees only their own pages. Everyone works from the same record and the same audit trail.

How AI is used

Numbers come from rules and models. AI explains and drafts. A person decides.

1. A person enterssymptoms, vitals, the conversation, the patient’s answers
2. Rules & model computeurgency level, drug checks, alerts, readmission risk: same input, same result
3. AI explains & draftssummaries, notes, questions, next actions, using only those results and the record
4. A person decidesconfirms, edits, signs, or rejects with a reason
5. Audit loginput, tools, output, model and decision for every step

How to test it

Play one patient through the whole journey. Sign in at /login with an email or phone; there is no password, a one-time code is sent instead. In this testing setup the code is always 1234.

1. Nurse · Arrivals. Sign in as Nurse Anna. Open Arrivals → + New patient. Enter a name, what the patient says (e.g. “Short of breath for 3 days, worse lying down, both ankles swollen”), the vitals, and an email for the patient app. Click Run AI check, then Confirm the level, or Change level with a reason.
2. Doctor · Consult. Sign out and sign in as Dr. Wilson. Open Consult and pick the patient. Talk using the microphone, or Type the conversation instead → Use this transcript. Click Draft note + suggest Rx, add the suggested medicines, resolve any Rx Check flag (or Override with reason), enter the diagnosis and Sign prescription.
3. Nurse · Follow-up. Sign in as the nurse again. Open Follow-up, pick the patient, review the AI-written questions, choose the days or send now, and approve. Send questions to patient sends them right away.
4. Patient · Questions and Chat. Sign in with the patient’s email. On Questions, answer like a patient who is getting worse (e.g. weight up 2 kg and ankles swollen, or “feeling worse”). In Chat, ask “When do I take my medicines?”.
5. Nurse or doctor · Alerts. Open Alerts: the rule that fired, the updated risk, Explain with AI, then Mark as handled or Dismiss with reason.
6. Check the trail. Audit log shows every step from 1 to 5. Insights (doctor, admin) shows risk bands, call outcomes and the model card. How it works walks through the same five steps inside the app.

Tips: staff can press ⌘K / Ctrl K to jump to any page or patient. The Briefing page opens with an AI summary of what to do first.

Watch the demo

A 5-minute walkthrough of one patient’s journey, showing each of the ten AI assistants at work. Recorded in the real app with synthetic data.